OpenAI Presents Research on Inference-Time Compute to Better AI SecurityMore inference-time compute reduces AI models' vulnerability to adversarial attacks.
Certain names make ChatGPT grind to a halt, and we know whyHard-coded filters can inadvertently disrupt usability and functionality in AI interactions, particularly for common names.AI tools face challenges from adversarial attacks that exploit system vulnerabilities, requiring ongoing evaluation and adjustment.
Why Cybercriminals Are Not Necessarily Embracing AI | HackerNoonAI aids malware detection but also introduces new cyber threats, demonstrated by threat actors using tools like ChatGPT.
Pentagon launches plan to keep its AI-powered tech from being hijackedAI systems vulnerable to adversarial attacks with visual 'noise' patches.Pentagon's GARD program works on identifying and defending against such vulnerabilities.
Why Cybercriminals Are Not Necessarily Embracing AI | HackerNoonAI aids malware detection but also introduces new cyber threats, demonstrated by threat actors using tools like ChatGPT.
Pentagon launches plan to keep its AI-powered tech from being hijackedAI systems vulnerable to adversarial attacks with visual 'noise' patches.Pentagon's GARD program works on identifying and defending against such vulnerabilities.
BEAST AI attack can break AI guardrails in 60 secondsEfficient adversarial attack phrases on LLMs developed by UMD computer scientists.BEAST technique for fast adversarial attacks requires Nvidia RTX A6000 GPU and minimal processing time.
Can AI Be Superhuman? Flaws in Top Gaming Bot Cast DoubtSuperhuman AI systems, like bots playing Go, can have vulnerabilities impacting safety and reliability.